https://www.janelia.org/sites/default/files/You%20%2B%20Janelia/Conferences/i2k2020-program.pdf
- Session 1: Monday, 30th November 15:00-19:00 UTC (16:00-20:00 CET) (10am-2pm ET)
- Session 2: Wednesday, 1st December 08:00-12:00 UTC (9:00-13:00 CET) (3am-7am ET)
#deep-learning-based-object-segmentation-and-img-restoration-stardist-and-csbdeep
- 0:00 Introduction of tutorial
- 0:10 Introduction of participants stating their motivation (1 slide, 1min in total)
- 0:40 Overview talk of CSBDeep and StarDist (Martin)
- 1:40 Dive into Tech talk (Uwe)
- 2:00 Break (15 mins)
- 2:15 Assignment to groups (breakout rooms)
- 2:30 Group work, tutorial notebooks
- 3:45 Wrap up
-
1 slide (1min in total) with an example image and the analysis problem that you want to solve (and that is relevant to this tutorial). You will be asked to share your screen at the beginning of the tutorial for 1 min to present this slide. (Note that the tutorial will be recorded. Please let us know if you don't want this and we will remove your section from the final recording.)
-
Prepare a Python environment, either
- (A) use your own system/environment (with a GPU)
- (B) create or activate your Google Colab account See below for detailed instructions
This section is intended for those who want a local installation of Python with CSBDeep and StarDist packages. Alternatively, you can run the example notebooks in the cloud via Binder (no GPU) or Google Colab (see below).
-
You need a C++ compiler to install StarDist. Please see this for details.
-
Create a new environment (with name
i2k-2020) via the provided environment files (see below).
(If you need help with managing conda environments, please see this guide.) -
Activate the new environment:
$ conda activate i2k-2020
If you have a CUDA-compatible GPU, try to install environment-gpu.yml:
$ conda env create -f environment-gpu.ymlIf this fails, you may try to manually install the specific versions of CUDA and cuDNN that are compatible with version 2.3.x of TensorFlow. Without this, computation will run on the CPU only. Then proceed with environment.yml:
$ conda remove --name i2k-2020 --all
$ conda env create -f environment.ymlThere is no GPU support for TensorFlow on macOS, hence you can only install environment.yml (and using gcc instead of the clang compiler, cf. this):
$ CC=gcc-10 CXX=g++-10 conda env create -f environment.ymlThe easiest way to get all example notebooks is by downloading a copy of the respective git repositories:
Make sure you have a Google account (https://accounts.google.com)
The tutorial notebooks are available here:
